# ExploitGym leaderboard

Agentic benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/agentic-exploit-gym

Every model the catalog carries a published ExploitGym value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Agentic |
| Measure | Working exploit generation |
| Tasks | 898 exploitation tasks |
| Difficulty | Advanced cybersecurity exploitation |
| Published by | [ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?](https://arxiv.org/abs/2605.11086) |

A controlled benchmark for evaluating whether AI agents can extend vulnerability-triggering inputs into working exploits.

## ExploitGym ranking

23 models with a published ExploitGym value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Working exploit generation | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Gemini 4 Carbon](https://modelscale.dev/models/gemini-4-carbon) | Google | 53.5 | Observed 2026-11-21 · source benchlm:models |
| 2 | [GPT-6 Cyber](https://modelscale.dev/models/gpt-6-cyber) | OpenAI | 48.5 | Observed 2026-11-21 · source benchlm:models |
| 3 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 42.4 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Gemini 4 Argon](https://modelscale.dev/models/gemini-4-argon) | Google | 39.5 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Claude Mythos 5.2](https://modelscale.dev/models/claude-mythos-5-2) | Anthropic | 38.1 | Observed 2026-11-21 · source benchlm:models |
| 6 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 35.1 | Observed 2026-11-21 · source benchlm:models |
| 7 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 33.7 | Observed 2026-11-21 · source benchlm:models |
| 8 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 33.3 | Observed 2026-11-21 · source benchlm:models |
| 9 | [GPT-5.6 Cyber](https://modelscale.dev/models/gpt-5-6-cyber) | OpenAI | 32.1 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Claude Mythos 5.1](https://modelscale.dev/models/claude-mythos-5-1) | Anthropic | 23.4 | Observed 2026-11-21 · source benchlm:models |
| 11 | [GPT-5.6 Terra](https://modelscale.dev/models/gpt-5-6-terra) | OpenAI | 23.2 | Observed 2026-11-21 · source benchlm:models |
| 12 | [GPT-6 Sol](https://modelscale.dev/models/gpt-6-sol) | OpenAI | 22.1 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Kimi K3.1 Pro](https://modelscale.dev/models/kimi-3-1-pro) | Moonshot AI | 20.7 | Observed 2026-11-21 · source benchlm:models |
| 14 | [MiMo-V2.6-Pro](https://modelscale.dev/models/mimo-v2-6-pro) | Xiaomi | 17.8 | Observed 2026-11-21 · source benchlm:models |
| 15 | [Claude Mythos Preview](https://modelscale.dev/models/claude-mythos-preview) | Anthropic | 17.5 | Observed 2026-11-21 · source benchlm:models |
| 16 | [DeepSeek V4.1 Flash 1017](https://modelscale.dev/models/deepseek-v4-1-flash) | DeepSeek | 15.3 | Observed 2026-11-21 · source benchlm:models |
| 17 | [GLM-5.3](https://modelscale.dev/models/glm-5-3) | Z.AI | 15.0 | Observed 2026-11-21 · source benchlm:models |
| 18 | [GPT-5.5](https://modelscale.dev/models/gpt-5-5) | OpenAI | 13.4 | Observed 2026-11-21 · source benchlm:models |
| 19 | [GPT-5.6 Luna](https://modelscale.dev/models/gpt-5-6-luna) | OpenAI | 12.4 | Observed 2026-11-21 · source benchlm:models |
| 20 | [GPT-6 Luna](https://modelscale.dev/models/gpt-6-luna) | OpenAI | 11.6 | Observed 2026-11-21 · source benchlm:models |
| 21 | [GPT-5.4](https://modelscale.dev/models/gpt-5-4) | OpenAI | 6.0 | Observed 2026-11-21 · source benchlm:models |
| 21 | [MiMo-V2.6-Flash](https://modelscale.dev/models/mimo-v2-6-flash) | Xiaomi | 6.0 | Observed 2026-11-21 · source benchlm:models |
| 23 | [Muse Spark 1.1](https://modelscale.dev/models/muse-spark-1-1) | Meta | 0.8 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
